As the number of greenfield and brownfield development opportunities dwindle, many operators are looking to refracture older existing wells for additional hydrocarbon production. While it is generally understood that a bigger second completions job will yield more oil production, influences from geology, previous fluid recovery, and inter-well spacing are less known. In this study, we use a series of machine learning models to disentangle the impacts of several well parameters and get an understanding of the conditions in which refracs produce more hydrocarbons.
We first separated the production data for refrac wells in the Williston basin into two groups, the time periods Before and After a refrac event occurred. Using only the Before time series data, we used a model to create a “no-refrac” forward projection. Next, we used the After production data points and subtract the “no-refrac” projection to quantify the Refrac Uplift in oil production. We built another ML model to predict the Refrac Uplift using a multitude of pre and post refrac features. Finally, Shapley values were calculated to estimate which features contribute the most to additional hydrocarbon production.
To no surprise, the additional proppant and fluid intensity pumped are two of most impactful features. While, initial proppant and fluid intensity are also high in importance ranking, we observe different trends between the initial and refrac sets of completions features. We also find that clay volume, water saturation, and porosity-height are the most important geological features. Furthermore, we examine the impacts of inter-well spacing and prior production before refrac.
Suboptimally-completed wells may have an abundance of oil trapped in the subsurface. As refracturing wells becomes a more common practice, engineers can use the workflows described in this paper to screen candidate wells more efficiently and design appropriate refracture completions jobs.